#Healthcare coverage prediction from 2009-2011 American Community Survey data
In February of this year, I took a programming test for a Chicago analytics company. It went poorly. The goal of the programming portion of the test was to supply probabilities of having healthcare coverage for the rows in the ACS dataset with missing coverage information (i.e. where hicov was NA). Several months later, I gave it a second shot!
###Directories: ####data
- HealcareData1.txt: ACS data used in the subsequent analysis. Alternatively, you can download from https://github.com/kjy/Healthcare/blob/master/HealthcareData1.txt
- frank_clean_data.txt: processed version of HealthcareData1.txt.
- train_data.csv: cleaned, scaled data with all hicov data available.
- test_data.csv: cleaned, scaled data with all hicov data as NA.
- hicov_probs.csv: predicted probabilities for yes/no hicov status for the rows present in test_data.csv.
####notebukz
- healthcare_demo_data_cleanup.ipynb : iPython notebook with a walkthrough of the data cleaning procedures. Note that I pull heavily from a similar analysis of this dataset done in R, which can be found at http://rstudio-pubs-static.s3.amazonaws.com/16587_f28745b007e744f784d5358c843a84e5.html
- healcare_analysis.ipynb: iPython notebook that loads the cleaned ACS data and tests several different types of models, along with brief descriptions of a few of the models. Generates hicov_probs.csv.
Special thanks to Jake for content and to Sarah for flexibility in scheduling!